Two of the big areas of improvement are using synonyms and similar words to expand search results, and automatic translation to find results in other languages.

These are generally useful, because they get more results from the pages we're after. Some examples from the article are adding matches for "song lyrics" when you search for "song words," and matches for "homicide" when you search for a string containing "murder."

But these techniques tend to foil the main ways I use Google when I'm doing research for a translation. When I use Google as a research tool for translation, I'm usually looking for exact phrases. I'm looking to see if a given English phrase is used in the same kinds of contexts as its Japanese equivalent.

I also use it to try and find the English names for Japanese organizations. If I've made a reasonable effort and no English name turns up, then I'll translate it myself, but if there's already an English equivalent with any currency, I think I have a duty to use that. Actually, when there is an English equivalent for an organization name, 95% of the time I find it on the organization's home page. But for the other 5%, it makes a big difference to be able to search on exact phrases.

In these cases, when Google gets too smart, it actually makes it harder to find what I'm after. I especially don't want Google to back-translate my search terms into Japanese, and show me Japanese matches!

You can kind of force Google into being more literal by prepending your search terms with a plus sign (+), but this doesn't always work, and the rules for when it works and when it doesn't are opaque (as far as I've been able to tell).

As Google gets increasingly smart about guessing what we really wanted to search for, I wonder if it will get increasingly hard for those of us who already know exactly what we're searching for, and just want to know if it exists.

On the other hand, though, if the scientists ever crack this mystery wide open (perhaps by giving up on computers with nothing but 0s and 1s to deal with and creating new machines that function more like a brain) then we’ll get our translating machine.

Well, IBM has created a cell-by-cell simulation of the human visual cortex. At 1.6 billion neurons, this is a huge leap from their simulation of a portion of a mouse cortex in 2006, and an entire rat cortex in 2007 (55 million neurons), so they seem to be making pretty rapid progress.

The main question seems to be whether Moore's Law will keep going until they get up to a complete human brain.